Intelligent water outlet control system and method for water purification equipment and water purification equipment

Abstract

The invention discloses an intelligent water outlet control system and method for water purification equipment and the water purification equipment, and relates to the technical field of intelligent control. The control system comprises a weight sensing module, a comprehensive acquisition module, a voice module and a central module; the weight sensing module and the central module share non-0 bearing weight; the comprehensive acquisition module acquires a distance data point set and environment data; the voice module induces and generates a voice signal; the central module processes the distance data point set, estimates the container capacity, determines the water outlet speed, the water outlet amount and the expected weight, and obtains the water outlet temperature by adopting a self-encoding voice recognition model in combination with a triggering effective strategy or predicts the water outlet temperature based on environmental data by adopting a support vector regression model based on the voice signal receiving condition within the waiting time; water discharging is executed according to the water discharging speed, the water discharging amount and the water discharging temperature, the state of the container is sensed based on the change of the bearing weight, water discharging is stopped when the container is abnormal or the bearing weight reaches the expected weight, and intelligent water discharging control over the water purification equipment is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control, and in particular to an intelligent water outlet control system, method and water purification equipment for water purification equipment. BACKGROUND

[0002] At present, the use of water purification equipment is gradually becoming routine. With the help of water purification equipment, users can directly use purified water sources.

[0003] The existing patent application with the publication number CN118605232A proposes a control method, device and storage medium for an energy-saving water purifier, which includes the following steps: obtaining the temporary identification and actual weight of the water receiving container; retrieving the stored identification in the water outlet database; determining whether the stored identification matches the temporary identification; if so, retrieving the theoretical weight corresponding to the stored identification from the water outlet database; otherwise, prompting the user to press the water outlet button; controlling the water outlet amount of the water purifier according to the difference between the theoretical weight and the actual weight, and prompting the customer after automatic water outlet; when the water outlet button is pressed, the water purifier dispenses water and simultaneously alarms, and asks the user whether to input the information of the water receiving container in this time into the water outlet database after the water receiving is completed, thereby realizing intelligent quantitative control of the water outlet amount of the water purifier and avoiding water resource waste.

[0004] However, the existing water outlet control of water purification equipment still relies on manual methods such as key pressing or wireless remote control to receive user instructions, and lacks a full-automatic water outlet control scheme integrating water outlet amount control, water outlet temperature control and water outlet anomaly monitoring. SUMMARY

[0005] The present application proposes an intelligent water outlet control system, method and water purification equipment for water purification equipment to address the deficiencies of the prior art, and realizes intelligent water outlet amount control, water outlet temperature control and water outlet anomaly monitoring through voice and automatic collaboration.

[0006] The technical solution adopted to achieve the purpose of the present application is as follows:

[0007] The intelligent water outlet control system for water purification equipment includes a weight sensing module, a comprehensive acquisition module, a voice module and a central module.

[0008] The weight sensing module senses the carrying weight When the carrying weight is not 0, the central module shares the carrying weight ;

[0009] The comprehensive acquisition module receives ranging signals or sensing signals, acquires distance data points or acquires environmental data, including environmental temperature , environmental humidity and environmental air pressure ;

[0010] The voice module performs vibration induction to generate voice signals when receiving the activation signal, performs voice prompting when receiving the exception signal or the completion signal, and stops vibration induction when receiving the silence signal ;

[0011] The central module receives the bearing weight , broadcasts the ranging signal and the activation signal and sets the waiting duration, receives the distance data point set , estimates the container capacity using a three-dimensional reconstruction volume estimation algorithm and determines the water outlet speed , the water outlet volume and the expected weight , based on the reception of the voice signal within the waiting duration, selects to use a self-encoding voice recognition model combined with a trigger effective strategy, analyzes the effectiveness of the voice signal to extract the water outlet temperature and broadcasts the silence signal or directly broadcasts the perception signal and the silence signal, receives environmental data and uses a support vector regression model to predict the water outlet temperature , executes water outlet according to the water outlet speed , the water outlet volume and the water outlet temperature , senses the container state based on the first-order gradient and the second-order gradient of the bearing weight , stops water outlet when the container is abnormal and broadcasts the exception signal, and continues water outlet until the expected weight is reached and broadcasts the completion signal.

[0012] Specifically, the weight sensing module continuously senses the bearing weight , when there is no container placed, the bearing weight is 0, when there is a container placed, the bearing weight becomes non-zero, and a data sharing link is established with the central module to share the bearing weight in real time , when the container leaves, the bearing weight returns to 0, and the data sharing link is disconnected.

[0013] Specifically, the comprehensive acquisition module includes a laser ranging unit, a temperature sensing unit, a humidity sensing unit and an air pressure sensing unit

[0014] The laser ranging unit receives the ranging signal, vertically projects a laser dot array to the water receiving platform , calculates the laser dot array based on the phase ranging principle The phase difference between each projected beam and its corresponding reflected beam is converted into a measurement distance, which is then calculated based on the projection beam's position on the laser array. Arrange the rows and columns correspondingly to generate a distance data point set. And provide feedback to the central module;

[0015] The temperature sensing unit receives the sensing signal and measures the forward voltage drop to obtain the ambient temperature. And provide feedback to the central module;

[0016] The humidity sensing unit receives the sensing signal and measures the capacitance of the humidity-sensitive medium to obtain the ambient humidity. And provide feedback to the central module;

[0017] The barometric pressure sensing unit receives the sensing signal and measures the voltage signal output by the Wheatstone bridge to convert it into ambient air pressure. And feedback is sent to the central module.

[0018] Specifically, the voice module receives an activation signal, continuously performs vibration sensing, generates an electrical signal by sensing the capacitance changes caused by sound wave vibrations in the environment, and then uses a filtering and amplification circuit to process and generate a voice signal. It also feeds back to the central module, which plays an abnormal or completion prompt voice when it receives an abnormal signal or a completion prompt voice, and stops vibration sensing when it receives a silence signal.

[0019] Furthermore, the central module includes a central control unit, a three-dimensional estimation unit, a voice recognition unit, and an automatic temperature control unit;

[0020] The central control unit receives the load weight. It establishes a data sharing link with the weight sensing module, broadcasts ranging and activation signals, sets a waiting time, and receives the container capacity. And proportionally converted to water output Combined water density Determine the expected weight Based on container capacity Determine the water output speed using fuzzy reasoning If the water temperature is obtained within the waiting time If a silent signal is broadcast and the water temperature is not obtained within the waiting time, the broadcast will be silent. Then, the broadcast sensing signal and the silence signal are based on the water output speed. Water output and outlet water temperature Perform water discharge, and continuously monitor the load-bearing weight during the discharge process. A weight gradient supervision strategy is employed to periodically calculate the first-order gradient. and second gradient The system senses the container's status; if an abnormality occurs, it stops discharging water and broadcasts an abnormality signal; otherwise, it continues discharging water until it reaches the required weight. equal to expected weight Then stop and broadcast a completion signal;

[0021] The three-dimensional estimation unit receives a set of distance data points. A three-dimensional reconstruction volume estimation algorithm is used to estimate the distance to the data point set. Reconstruct a 3D model of the container and estimate its capacity. And feedback is given to the central control unit;

[0022] The speech recognition unit receives speech signals. And process to generate speech features The input is an autoencoder speech recognition model, which combines local features extracted by a combined convolutional network with global features extracted by a Transformer encoder to generate speech codes. The CTC decoder is used to assist the Transformer decoder in learning speech coding. Alignment with Chinese characters to generate speech content. Identify valid speech content based on trigger-based effective strategies. And extract water temperature Feedback to the central control unit;

[0023] The automatic temperature control unit receives environmental data, maps it to a high-dimensional space using a support vector regression model, and performs linear regression to predict the outlet water temperature. And feedback is given to the central control unit.

[0024] Specifically, the central control unit receives the container capacity. According to the water output ratio With container capacity The product determines the water output. Combined water density The calculated water output is water weight Determine the desired weight equal to the load-bearing weight With the weight of water The sum of.

[0025] Furthermore, we establish container fuzzy sets, including small... ,middle ,big ,Small ,middle ,big These are used to describe different ranges of container capacity using fuzzy descriptions. Using triangular membership functions respectively , , Describe it;

[0026] Establish a fuzzy rule table reflecting the correspondence between the container fuzzy set, the water outlet velocity fuzzy set, and the speed setting. The water outlet velocity fuzzy set includes low... ,middle and high ,Low ,middle and high These are used to describe different ranges of water velocities with varying degrees of fuzziness. Gear speeds include , and The fuzzy rule table is specifically as follows: small Corresponding low correspond ,middle Corresponding correspond ,big Corresponding high correspond ;

[0027] Container capacity Substitute the values ​​into the membership function of the triangle to calculate the membership degree. , , Water output speed It is a weighted sum of membership degree and corresponding gear speed.

[0028] Furthermore, the weight gradient monitoring strategy, during the effluent process, is based on monitoring intervals. Periodically set up monitoring nodes to obtain the first... Each monitoring node load-bearing weight Calculate the first Each monitoring node first gradient With water output speed The difference and second gradient , respectively, as the first-order gradient difference and the difference between the second and third gradients Perform node anomaly detection and determine the first-order gradient difference. Difference with second gradient Whether each is less than the corresponding fluctuation threshold, if and only if the first-order gradient difference and the difference between the second and third gradients When all are less than the corresponding fluctuation threshold, determine the first Each monitoring node For normal node, the rest are determined as abnormal nodes, whether the number of continuous abnormal nodes is greater than or equal to 3 is determined, if less than 3, the node abnormality determination is continued at the next supervision node until the water is stopped, if greater than or equal to 3, the container is determined to be abnormal, the water is immediately stopped and the abnormal signal is broadcasted.

[0029] Further, the distance data point set is converted into a three-dimensional model of the container and the capacity of the container is estimated by using a three-dimensional reconstruction volume estimation algorithm The specific steps include:

[0030] The water outlet axis is taken as the axis, the intersection of the axis and the water receiving platform is taken as the origin, and the two straight lines in the plane of the water receiving platform passing through the origin and parallel to the wide side and the long side of the water receiving platform are taken as the axis and the axis respectively, a space coordinate system is constructed and the distance data point set is placed into, and a point cloud data set is generated ;

[0031] The point cloud data set is traversed , the neighborhood of the first target point and the first neighbor point are determined by using the neighbor algorithm, the local surface with the smallest distance to the first neighbor point is fitted by using the least square method, the local surface of all target points in the point cloud data set is obtained, and a preliminary three-dimensional model is constructed , is the total number of neighbor points

[0032] The partial derivatives of the first target point in the three coordinate axis directions are calculated to construct a gradient vector, if the modulus of the gradient vector of the target point is greater than or equal to a gradient threshold value, the target point is determined to be a container edge point, all container edge points in the point cloud data set are obtained, the local surface of the non-edge points in the preliminary three-dimensional model is removed, and a preliminary container three-dimensional model is generated

[0033] The covariance matrix is calculated according to the first neighbor point of the first target point and eigenvalue decomposition is performed, three eigenvalues are obtained, and the ratio of the smallest eigenvalue to the sum of the three eigenvalues is taken as the first target point ​​​​​​​​​​​​​​​​​​The curvature is calculated and compared with a curvature threshold. If the curvature is greater than or equal to the curvature threshold, it is determined to be a corner point, and a point cloud dataset is obtained. All corner points;

[0034] The Laplacian smoothing algorithm is used to calculate the average neighborhood coordinates of all nearest neighbors in the neighborhood of each corner point. The smoothing factor is used to control the offset of the corner point towards the average neighborhood coordinates to smooth the surface of the initial container 3D model and generate the container 3D model.

[0035] Estimating container capacity using the infinitesimal integral method or the proportional calculation method. The infinitesimal integral method, based on the concept of infinitesimal elements, divides the three-dimensional model of a container into multiple cuboids, and then... Regarding the 3D model of the container Axis range and The double integral within the axis range as the container capacity The proportional calculation method generates a large number of particles within a minimum directed bounding box and calculates the percentage of particles within that box. The product of the minimum directed bounding box volume and the percentage of particles within that box is taken as the container capacity. .

[0036] Furthermore, preprocessing speech signals It also uses a self-encoding speech recognition model to parse and generate speech content. The specific steps include the following:

[0037] The speech signal is processed by a first-order high-pass filter. Pre-emphasis is applied, and the signal is divided into multiple segments with a fixed frame length. Each segment is multiplied by a Hamming window function to achieve windowing smoothing. The signal is then transformed to the frequency domain using Fourier transform. A Mel filter bank is constructed based on the conversion relationship between Mel frequency and linear frequency, and then applied to the speech signal. The speech features are obtained by performing frequency domain convolution on the spectral information and taking the logarithm. ;

[0038] Convolutional encoders in autoencoder speech recognition models process speech features. The convolutional encoder includes a first feedforward network. Multi-head attention network Combinatorial convolutional networks Second feedforward network Speech codes are generated by combining macaron structures. Among them, combined convolutional networks Includes the first pointwise convolutional layer, gated linear layer, channel-wise convolutional layer, The activation function and the second pointwise convolutional layer are used. The pointwise convolutional layer is used to extract channel information, the gated linear layer is used to filter features, and the channelwise convolutional layer is used to extract local feature information in the time dimension. The activation function is used to increase the nonlinear expressive power. The macaron structure refers to superimposing the network output and the network input according to a certain ratio and using it as the input of the next network.

[0039] The joint encoder of the autoencoder speech recognition model parses the speech code. A CTC decoder is introduced into the Transformer decoder. The CTC decoder includes a linear layer and... Activation function, used to establish speech coding The probability distribution of alignment paths to Chinese character sequences; the Transformer decoder uses a self-attention mechanism to encode speech. The content is converted into Chinese characters, and the order of the Chinese characters is adjusted according to the probability distribution of the alignment path to generate speech content. .

[0040] Specifically, effective triggering strategies are used to determine the content of speech. The effectiveness of the algorithm is assessed by using automata algorithms to match speech content based on keyword regular expressions. If keywords are present, the audio content is determined. Effective; the system uses an automaton algorithm to extract speech content based on temperature regular expression matching. The outlet water temperature If it does not exist, then determine the voice content. invalid.

[0041] Furthermore, the automaton algorithm employs a deterministic finite automaton, defining a finite set of states based on regular expressions, including an initial state and an accepting state, corresponding to an empty string and a regular expression, respectively. Starting from the initial state, it processes the speech content... The system reads the text character by character, uses the current state and the state transition function of the current character input to generate the next state, and then reads the entire speech content. Then, if it is determined that the finite automaton is in the receiving state, then the speech content... It matches the regular expression successfully; otherwise, it fails to match.

[0042] Furthermore, the support vector regression model needs to be pre-trained, which includes the following specific steps:

[0043] Collection includes Training sample set of group training samples , No. Group training samples Including ambient temperature Ambient humidity Ambient air pressure and outlet water temperature , , where N is the total number of training samples;

[0044] where w is the weight vector and b is the bias term of the model; , where x is the input vector, and the insensitive loss function is used to measure the error between the model prediction and the true value; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group;

[0045] where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group;

[0046] where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group;

[0047] where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group;

[0048] where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; where e is the error of the i-th training sample in the i-th group; Take partial derivatives and set them equal to zero to obtain the dual problem and solve it using the sequential minimal optimization algorithm to obtain the optimal first multiplier vector and the optimal second multiplier vector ;

[0049] Substitute the optimal first multiplier vector and the optimal second multiplier vector back into the Lagrangian function with respect to the weight vector and the bias term , set the partial derivatives equal to zero and solve to obtain the optimal weight vector and the optimal bias term , obtaining the optimal model representation , pre-training is complete.

[0050] The intelligent water outlet control method for water purification equipment is realized based on the intelligent water outlet control system for water purification equipment, and includes the following specific steps:

[0051] Obtain a non-zero bearing weight , collect distance data points and estimate the container capacity by a three-dimensional reconstruction volume estimation algorithm , determine the water outlet quantity according to the water outlet proportion and calculate the expected weight , determine the water outlet speed based on fuzzy reasoning ;

[0052] Set a waiting time, execute vibration sensing to capture sound waves in the environment and convert them into voice signals ;

[0053] If a voice signal is obtained within the waiting time , use a self-encoding voice recognition model to analyze the pre-processed voice signal to generate voice content , based on the trigger effective strategy, use the automatic machine algorithm combined with regular expressions to judge the effectiveness of the voice content , when the voice content is valid, extract the water outlet temperature and stop vibration sensing, when the voice content is invalid, continue to wait for a new voice signal until the waiting time ends;

[0054] If no voice signal is obtained within the waiting time or the obtained voice signal is invalid, collect the ambient temperature , ambient humidity and ambient air pressure , the outflow temperature is predicted by a support vector regression model ;

[0055] According to the outflow speed , the outflow volume and the outflow temperature , an outflow and weight gradient supervision strategy is executed to monitor the carrying weight in real time and periodically judge the volatility of the first-order gradient and the second-order gradient to sense whether the container is abnormal, stop the outflow and play an abnormal prompt voice when the container is abnormal, and continue the outflow until the carrying weight equals the expected weight , then stop the outflow and play a completion prompt voice.

[0056] A water purification device for carrying the intelligent outflow control system for the water purification device, comprising a water purifier, a gravity sensor, a laser range finder, a temperature sensor, a humidity sensor, an air pressure sensor, an integrated voice device and a single-chip microcomputer;

[0057] The water purifier is controlled by the single-chip microcomputer to execute outflow at the outflow speed , the outflow volume and the outflow temperature ;

[0058] The gravity sensor is installed directly below the water receiving platform of the water purifier to continuously sense the carrying weight of the water receiving platform , and share the carrying weight with the single-chip microcomputer when the carrying weight is not 0 ;

[0059] The laser range finder is installed directly above the water receiving platform of the water purifier and is controlled by the single-chip microcomputer to vertically project a laser dot matrix to the water receiving platform to feed back a distance data point set to the single-chip microcomputer;

[0060] The temperature sensor is controlled by the single-chip microcomputer to obtain the ambient temperature and feed back to the single-chip microcomputer;

[0061] The humidity sensor is controlled by the single-chip microcomputer to obtain the ambient humidity and feed back to the single-chip microcomputer;

[0062] The air pressure sensor is controlled by the single-chip microcomputer to obtain the ambient air pressure and feed back to the single-chip microcomputer;

[0063] The integrated voice device is controlled by the single-chip microcomputer to execute vibration sensing to capture sound waves in the environment, generate voice signals and feed back to the single-chip microcomputer or play abnormal prompt voice and completion prompt voice;

[0064] The single-chip microcomputer controls the laser range finder, temperature sensor, humidity sensor, air pressure sensor and integrated voice device to perform collection or playing, and determines the water speed, water amount and water temperature based on the feedback data . . . controls the water outlet of the water purifier, judges the abnormality of the container by monitoring the shared bearing weight of the gravity sensor, and stops the water outlet after the container is abnormal or the water outlet is completed.

[0065] Compared with the prior art, the present application has the following advantages:

[0066] The distance data point set is collected by the laser ranging device, the volume of the container is estimated by using the three-dimensional reconstruction volume estimation algorithm, the water amount is determined according to the water outlet proportion, the expected weight is calculated, the water speed is determined based on fuzzy reasoning, the water outlet is automatically stopped when the bearing weight reaches the expected weight, the automation of the water outlet amount of the water purifier is realized, and the problems of the container being full and difficult to take and the water being splashed and wasted due to too fast water outlet are avoided;

[0067] The voice signal is captured by the integrated voice device, the voice content is generated by using the self-encoding voice recognition model, the effective voice content is extracted to determine the water temperature based on the trigger effective strategy, the standby scheme is set, the environment temperature, environment humidity and environment air pressure are collected when no voice signal is received or all the received voice signals are invalid within the waiting time, and the water temperature suitable for the current environment is obtained by using the support vector regression model, so that the voice control and automatic control of the water outlet temperature of the water purifier are realized;

[0068] The placement of the container is automatically determined by the change of the bearing weight, the first-order gradient and second-order gradient fluctuation are monitored in real time during the water outlet process by using the weight gradient supervision strategy to automatically judge the abnormal state of the container and give a prompt, and the water outlet abnormality monitoring of the water purifier is realized. BRIEF DESCRIPTION OF DRAWINGS

[0069] Figure 1 It is a schematic diagram of the intelligent water outlet control system for the water purifier in the present application;

[0070] Figure 2 It is a self-encoding voice recognition model diagram in the present application;

[0071] Figure 3 It is a flow chart of the intelligent water outlet control method for the water purifier in the present application;

[0072] Figure 4 It is a schematic diagram of the water purifier in the present application.

[0073] Reference numerals: 1. Water purifier; 2. Gravity sensor; 3. Laser rangefinder; 4. Temperature sensor; 5. Humidity sensor; 6. Barometric pressure sensor; 7. Integrated voice device; 71. MEMS microphone; 72. Speaker; 8. Microcontroller. Detailed Implementation

[0074] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0075] Example 1

[0076] like Figure 1 As shown, a specific embodiment of the present invention discloses an intelligent water output control system for a water purification device, including a weight sensing module, a comprehensive acquisition module, a voice module, and a central module;

[0077] Weight sensing module senses the load weight Changes and bearing weight When the load is not zero, the load capacity is shared in real time with the central module. ;

[0078] When the integrated acquisition module receives the ranging signal, it projects a laser dot array to collect a set of distance data points. The central module receives the sensing signals and collects the ambient temperature data. Ambient humidity and ambient air pressure And together with the feedback to the central module;

[0079] The voice module receives an activation signal, performs vibration sensing, and generates a voice signal upon detecting sound wave vibrations. It also feeds back to the central module. If it receives an abnormal signal or a completion signal, it will provide a prompt based on the corresponding built-in voice. If it receives a silence signal, it will immediately stop the vibration sensing.

[0080] The central module receives the load weight. Broadcast ranging and activation signals and set the waiting time, then receive distance data point sets. The container capacity is estimated using a three-dimensional reconstruction volume estimation algorithm. And adaptively determine the water output velocity Water output and expected weight Based on voice signal during the waiting time Based on the reception status, a self-encoding speech recognition model combined with a trigger-effective strategy is selected to parse the speech signal. The effectiveness depends on the temperature of the extracted water. It broadcasts a silence signal or selects to directly and synchronously broadcast a sensing signal and a silence signal, and receives the ambient temperature. Ambient humidity and ambient air pressure The effluent temperature is predicted using a support vector regression model. According to the water output speed Water output and outlet water temperature Water discharge is performed, and the discharge process is based on the load-bearing weight. first gradient and second gradient If the sensor detects an abnormality in the container, it will stop discharging water and broadcast an abnormality signal if the container is malfunctioning. If the container is functioning normally, it will continue discharging water until it reaches the required weight. equal to expected weight Then the water flow stops and a completion signal is broadcast.

[0081] Specifically, the weight sensing module is based on a gravity sensor and continuously senses the load-bearing weight of the water receiving platform. When the water receiving platform has no container placed on it, the load-bearing weight The load-bearing capacity is always 0 when the container is placed on the water receiving platform. When the value becomes non-zero, the weight sensing module establishes a data sharing link with the central module, and shares the load-bearing weight with the central module in real time. When the container leaves the water receiving platform, the load-bearing weight Returning to zero, the established data sharing link is disconnected.

[0082] Specifically, the integrated data acquisition module includes a laser ranging unit, a temperature sensing unit, a humidity sensing unit, and a barometric pressure sensing unit.

[0083] The laser ranging unit is based on a laser rangefinder, which receives ranging signals and vertically projects a laser dot array onto the water receiving platform. Laser dot matrix The Middle Line 1 Column of projected beams A corresponding reflected beam is generated upon contact with the water receiving platform or container. Based on the principle of phase ranging, the projected beam is With reflected beam phase difference Convert to distance measurement The specific formula for phase ranging is as follows:

[0084] ;

[0085] in, For laser dot matrix The wavelength will be used to measure distance. Based on the projected beam In laser dot array number of rows Number of columns corresponding arrangement, generating distance data points set and feedback hub module;

[0086] The temperature sensing unit is realized based on a temperature sensor, receives a sensing signal, measures the forward voltage drop of the semiconductor PN junction in real time, and obtains the ambient temperature based on the linear correlation between the forward voltage drop and the ambient temperature and feedback hub module, wherein the ambient temperature of the semiconductor PN junction decreases by 2mV per 1℃ increase;

[0087] The humidity sensing unit is realized based on a humidity sensor, receives a sensing signal, measures the capacitance of the humidity-sensitive medium between the parallel plate metal electrodes in real time, obtains the dielectric constant of the humidity-sensitive medium by dividing the product of the capacitance and the electrode distance by the parallel plate area based on the parallel plate capacitance formula, and obtains the ambient humidity through a polynomial model and feedback hub module, wherein the humidity-sensitive medium can absorb water molecules in the environment, and when the ambient humidity increases, the dielectric constant of the humidity-sensitive medium also increases, and the polynomial model is determined by fitting a plurality of humidity-dielectric constant pairs calibrated through experiments;

[0088] The air pressure sensing unit is realized based on an air pressure sensor, receives a sensing signal, measures the voltage signal output by the Wheatstone bridge in real time, and converts the voltage signal into ambient air pressure according to the proportion calibrated through experiments and feedback hub module, wherein the resistance change of the pressure-sensitive resistor in the air pressure sensor is positively correlated with the change of the ambient air pressure When the resistance value changes, the balance state of the Wheatstone bridge is broken, and a voltage signal proportional to the ambient air pressure is generated.

[0089] Specifically, the voice module is realized based on an integrated voice device, receives an activation signal to enter an activated state, continuously performs vibration sensing through a MEMS microphone, when there is a sound wave in the environment, the sound wave kinetic energy causes the diaphragm to vibrate to change the capacitance, since the capacitance is externally connected to a constant power supply, the charge quantity remains unchanged, the capacitance change causes the voltage change to generate an electric signal, which is processed by a filter and amplifier circuit to generate a voice signal and feedback hub module, when receiving an abnormal signal or a completion signal, the built-in abnormal prompt voice or completion prompt voice is extracted and played through a loudspeaker, when receiving a silence signal, the vibration sensing of the MEMS microphone is immediately stopped and enters a silent state.

[0090] Further, the hub module includes a total control unit, a three-dimensional estimation unit, a voice recognition unit, and an automatic temperature setting unit;

[0091] The total control unit is realized based on a single-chip microcomputer, receives a bearing weight​ And establish a data sharing link with the weight sensing module, broadcast the ranging signal and the activation signal and set the waiting time, receive the container capacity And proportionally converted into water output Further combined with the bearing weight And water density Determine the expected weight , the container capacity Attributed to the container fuzzy set to infer to determine the water speed , if the water temperature is obtained within the waiting time Then broadcast the silence signal, if the water temperature is not obtained within the waiting time Then broadcast the sensing signal and the silence signal synchronously, based on the water speed , water output And water temperature Control the water output of the water purification equipment, continuously monitor the bearing weight based on the data sharing link during the water output process , adopt the weight gradient supervision strategy to supervise whether the container is abnormal based on the first order gradient And the second order gradient Of the bearing weight , stop water when the container is abnormal and broadcast the abnormal signal, and continue to output water until the bearing weight Equal to the expected weight Then stop the water and broadcast the completion signal

[0092] The three-dimensional estimation unit receives the distance data point set , converts the distance data point set Into a three-dimensional model of the container using a three-dimensional reconstruction volume estimation algorithm, estimates the container capacity And feedback to the master control unit

[0093] The speech recognition unit receives the speech signal And converts the speech features after preprocessing , input the self-encoding speech recognition model, the self-encoding speech recognition model is improved based on the Transformer model by introducing a combination convolution network and a CTC decoder, the combination convolution network extracts local features of speech features And the global features extracted by the Transformer encoder are embedded with each other to generate speech encoding , introduce the CTC decoder in the Transformer decoder, the CTC decoder converts the speech encoding Into a sequence of Chinese characters to assist the Transformer decoder in learning the alignment relationship between the speech encoding And Chinese characters, generate speech content Determine speech content based on effective triggering strategies. Validity to extract valid speech content The outlet water temperature And it feeds back to the central control unit, where the autoencoded speech recognition model is as follows: Figure 2 As shown;

[0094] The automatic temperature control unit receives the ambient temperature. Ambient humidity and ambient air pressure A support vector regression model is used to map the water to a high-dimensional space, and linear regression is performed in the high-dimensional space to predict the outlet water temperature. And feedback is given to the central control unit.

[0095] Specifically, the central control unit receives the container capacity. According to the water output ratio Container capacity Converted to water output To avoid the problem of difficulty in retrieving an overflowing container, due to the density of water... The water output is water weight At this time, the water purification equipment has not yet dispensed water, and the load-bearing weight... Equal to the weight of the container, therefore, the expected weight upon completion of water discharge. equal to the load-bearing weight With the weight of water sum.

[0096] Furthermore, container fuzzy sets include small ,middle ,big ,Small ,middle ,big These are used to describe different ranges of container capacity using fuzzy descriptions. Using triangular membership functions respectively , , The description is as follows:

[0097] ;

[0098] ;

[0099] ;

[0100] in, , , and Variables Range segmentation, variable Uniformly refers to container capacity Any value within the range, in this embodiment, is taken as 0.5, 1, 2, and 3.5, respectively, in liters;

[0101] A fuzzy rule table reflecting the correspondence between the container fuzzy set, the water outlet speed fuzzy set, and the gear speed is established, as shown in Table 1, and the details are as follows:

[0102] Table 1 Fuzzy Rule Table

[0103]

[0104] in, , and The low values ​​in the fuzzy set of water output velocity are respectively ,middle ,high The corresponding gear speed, low ,middle ,high These are used to describe different ranges of water velocities with varying degrees of fuzziness. , container capacity Substitute into the membership function of the triangle respectively , , To calculate membership degree , , Then the water output speed It is the weighted sum of membership degree and corresponding gear speed, i.e. In this embodiment, , and Take values ​​of 0.1, 0.2, and 0.4 respectively, in liters per second.

[0105] Furthermore, a weight gradient supervision strategy is adopted based on the load weight. first gradient and second gradient Is the fluctuation sensor container abnormal, and what is its load-bearing weight? first gradient Reflects the load-bearing weight Regarding the rate of change over time, since the water purification equipment operates at the water output speed... A constant outflow rate, under normal conditions, with a first-order gradient. The water output speed should always be maintained. Small fluctuations in the vicinity, bearing weight second gradient Reflecting the first-order gradient Regarding the rate of change over time, i.e., the outflow velocity the change speed of the weight of the node, when no abnormal condition occurs, the second-order gradient should always fluctuate in a small range around 0, the weight gradient supervision strategy is based on the monitoring interval periodically sets a supervision node, obtains the weight of the supervision node , calculates the first-order gradient and the second-order gradient of the supervision node , and the specific calculation formula is as follows:

[0106] ;

[0107] ;

[0108] wherein, , and are the weight, the first-order gradient and the second-order gradient of the supervision node , the difference between the first-order gradient of the supervision node and the water outlet speed is taken as the first-order gradient difference value , and the second-order gradient of the supervision node is directly taken as the second-order gradient difference value ;

[0109] the node abnormality judgment is performed to judge whether the first-order gradient difference value and the second-order gradient difference value of the supervision node are respectively less than the corresponding first-order fluctuation threshold and second-order fluctuation threshold, and only when the first-order gradient difference value is less than the first-order fluctuation threshold and the second-order gradient difference value is less than the second-order fluctuation threshold, the supervision node is determined to be a normal node, otherwise, the supervision node is determined to be an abnormal node, and it is judged whether the number of continuous abnormal nodes is greater than or equal to 3, if less than 3, the node abnormality judgment is continued at the supervision node until the water outlet is stopped, and if greater than or equal to 3, the container is determined to be abnormal, the water outlet is immediately stopped, and an abnormal signal is broadcasted.

[0110] ​​​​​​​​​Furthermore, a 3D reconstruction volume estimation algorithm is used to estimate the distance to the data point set. Convert into a 3D model of the container and estimate its capacity. The specific steps include the following:

[0111] Using the water outlet axis of the water purification equipment as Axis, with The intersection of the axis and the water receiving platform is taken as the origin. Two straight lines in the plane containing the water receiving platform, passing through the origin and parallel to the width and length of the platform respectively, are taken as... shaft and Axis, construct a spatial coordinate system, obtain the height of the laser rangefinder relative to the water receiving platform, and subtract the distance data point set. The distance to each data point is obtained, and the coordinates of each data point in the spatial coordinate system are calculated. The axis value determines the projection of the laser rangefinder center onto the water-receiving platform. Axis values ​​and The axis value is calculated based on the spacing of the laser dot array, determining the spatial coordinate system of each data point. Axis values ​​and Axis values ​​represent the distance from the data point set. Each data point is transformed into a point cloud in a spatial coordinate system, generating a point cloud dataset. ;

[0112] Using moving least squares method on point cloud datasets Surface fitting is performed by traversing the point cloud dataset based on the idea of ​​local weighted least squares. For the first Target points ,pass Nearest neighbor algorithm to determine target point neighborhood And obtain the neighborhood within Given a set of nearest neighbors, fit a local surface using the least squares method to make the neighborhood... Inside To obtain the point cloud dataset, minimize the Euclidean distance from the nearest neighbor points to the local surface. A preliminary 3D model is constructed using local surfaces at all target points. This represents the total number of nearest neighbors.

[0113] For point cloud datasets The first in Target points The gradient-based edge extraction method is used to calculate the target points respectively. exist axis, shaft and The partial derivatives along the axes are combined to construct the gradient vector, and the magnitude of the gradient vector is compared with the gradient threshold to determine the target point. Whether it is a container edge point or not, if the magnitude of the gradient vector is greater than or equal to the gradient threshold, then it is determined as a target point. For points on the container edge, if the magnitude of the gradient vector is less than the gradient threshold, then the target point is determined. For non-edge points, obtain the point cloud dataset. By identifying all container edge points and removing local surfaces that are not edge points from the preliminary 3D model, a preliminary 3D model of the container can be obtained.

[0114] For point cloud datasets The first in Target points Based on neighborhood Inside Calculate the covariance matrix for each nearest neighbor. The specific form is as follows:

[0115] ;

[0116] in, Representing the neighborhood Inside Nearest neighbor points about shaft and The covariance of the axis, The specific calculation formula is as follows:

[0117] ;

[0118] in, and Neighborhood Inner Neighboring points of Axis values ​​and Axis value, and Neighborhood Inside Nearest neighbor Axis Mean and Axis mean, ;

[0119] For the Target points covariance matrix Perform eigenvalue decomposition to obtain information about... axis, shaft and Calculate the 3 eigenvalues ​​of the axis. Target points The curvature is the ratio of the smallest eigenvalue among the three eigenvalues ​​to the sum of the three eigenvalues, and is further compared with the curvature threshold. If the curvature is... Target points If the curvature is greater than or equal to the curvature threshold, then the first... Target points Let be a corner point, if the first... Target points If the curvature is less than the curvature threshold, then the first... Target points For non-corner points, obtain the point cloud dataset. All corner points;

[0120] The Laplace smoothing algorithm is used to process the preliminary container 3D model. For each corner point in the preliminary container 3D model, the neighborhood average coordinates of all nearest neighbor points in the neighborhood of the corner point are calculated. The smoothing factor controls the corner point to shift towards the neighborhood average coordinates. The shift distance is equal to the error distance between the corner point and the neighborhood average coordinates multiplied by the smoothing factor, which reduces the noise and irregularity on the surface of the preliminary container 3D model and generates the container 3D model.

[0121] Estimating container capacity using the infinitesimal integral method or proportional calculation method combined with a three-dimensional model of the container. Integral method of infinitesimal elements The planar model generates a micro-mesh and divides the 3D model of the container into multiple cuboids, each with a width and length of [missing information]. and The height of the cuboid is the height of the top and bottom surfaces of each micro-mesh of the container 3D model. Axis height difference Because the micro-mesh is small enough, the container capacity is... It can be approximated as the sum of the volumes of each cuboid, i.e., the container capacity. The height of the cuboid Regarding the 3D model of the container Axis range and The minimum directed bounding box of the container's 3D model is obtained by using a double integral within the axis and a proportional calculation method. The minimum directed bounding box is a cuboid that can completely contain the container's 3D model and has the smallest volume. A large number of particles are randomly generated within the minimum directed bounding box. The total number of particles within the container's 3D model is counted and its ratio to the total number of particles is calculated to obtain the proportion of particles within the container. Therefore, the container capacity is obtained. It is the product of the minimum directed bounding box volume and the percentage of particles inside it.

[0122] Furthermore, preprocessing speech signals It also uses a self-encoding speech recognition model to parse and generate speech content. The specific steps include the following:

[0123] The speech signal is pre-emphasized to enhance high-frequency energy ; The speech signal is a non-stationary signal that changes over time, but can be approximated as a stationary signal in a short time, and the speech signal is divided into frames with a fixed frame length , and to avoid discontinuity in the framed signal, each frame of the framed signal is multiplied by a Hamming window function to implement windowing of the framed signal, so that the frame boundary can be smoothly transitioned, and the windowed framed signal is converted to the frequency domain by Fourier transform to obtain spectral information of the speech signal ; The Mel filter bank is constructed based on the conversion relationship between Mel frequency and linear frequency, and is convolved with the spectral information of the speech signal ;

[0124] The convolutional encoder in the auto-encoding speech recognition model processes the speech feature , the convolutional encoder is improved by a Transformer encoder, including a first feedforward network , a multi-head attention network , a combined convolutional network and a second feedforward network , combined through a macaron structure to generate speech encoding , the macaron structure refers to the output of the network is added according to a certain proportion and the input of the network to generate an overlapping output, and the overlapping output is taken as the input of the next network, and the mathematical expression is as follows:

[0125] ;

[0126] ;

[0127] ;

[0128] ;

[0129] wherein, is the first overlapping output generated by adding half of the output of the first feedforward network and the speech feature , is the second overlapping output generated by adding the output of the multi-head attention network and the first overlapping output , is the third overlapping output generated by adding the output of the combined convolutional network and the second overlapping output , For batch normalization operation, the first feedforward network , the multi-head attention network and the second feedforward network For the existing structure of the Transformer encoder, the combined convolutional network is not a traditional single convolutional layer, including a first pointwise convolutional layer, a gated linear layer, a channel-wise convolutional layer, an activation function and a second pointwise convolutional layer, the pointwise convolutional layer is used to extract channel information in the feature dimension, the gated linear layer is used to filter speech features, the channel-wise convolutional layer is used to extract local feature information of speech in the time dimension, the activation function is used to increase the nonlinear expression ability of speech features, the introduction of the combined convolutional network effectively makes up for the defect that the local feature capturing ability of the Transformer encoder is weak;

[0130] The joint encoder of the auto-encoding speech recognition model parses speech encoding , a CTC decoder is introduced on the basis of the Transformer decoder, the CTC decoder includes a linear layer and an activation function, which is used to establish a probability distribution of the alignment path of speech encoding to Chinese character sequence, the Transformer decoder converts speech encoding into accurate Chinese character content based on the self-attention mechanism, and adjusts the arrangement order of Chinese characters in the Chinese character content to generate speech content .

[0131] Specifically, the trigger effective strategy is used to judge the effectiveness of the speech content to avoid misrecognition caused by user daily conversation in the use of water purification equipment, and an automaton algorithm is used to match whether there is a keyword in the speech content according to the keyword regular expression, if there is a keyword, it is determined that the speech content is valid, and an automaton algorithm is used to match and extract the outlet water temperature in the speech content according to the temperature regular expression, if there is no keyword, it is determined that the speech content is invalid, and this speech signal is ignored .

[0132] Further, the automaton algorithm uses a deterministic finite automaton, which defines a finite state set based on a regular expression, including an initial state and an accepting state. The state is essentially a string, the initial state is an empty string, and the accepting state is a text corresponding to the regular expression. In this embodiment, the accepting state includes the keyword and the outlet water temperature , when determining the matching of a finite automaton, starting from the initial state, read each character of the speech content one by one. Through the state transition function, determine the next state based on the current state and the currently read character. And the state transition of the finite automaton is deterministic, that is, each state has a uniquely determined transition state for a specific read character. If the speech content is completely read and the finite automaton is in the accepting state, then the speech content matches the regular expression represented by the finite automaton, and in other cases, the matching fails.

[0133] Specifically, the temperature regular expression is: "\d+(\.\d+)? degrees", where "\d+" is used to match a single digit or multiple consecutive digits in the speech content , for example , , , "(\.\d+)" is a grouping used to match the situation where at least one digit is connected after the decimal point in the speech content , for example , , . "?" indicates that the grouping "(\.\d+)" is non-essential, that is, after the single digit or multiple consecutive digits corresponding to "\d+", "degrees" can also be directly connected. "Degrees" is used to match "degrees" in the speech content . The keyword regular expression is defined according to different keywords. To avoid the inconvenience of user voice control caused by strict keyword limitations, generally, the keyword regular expression will include synonyms replacement of some content in the keyword. Taking the keyword "Hello, Youyou" as an example, the keyword regular expression can be set as "(you (good|your)|hello), Youyou", indicating that in the speech content , "Hello, Youyou", "Hello, Youyou", and "Hello, Youyou" can all match successfully.

[0134] Furthermore, using the support vector regression model requires pre-training to obtain the optimal model representation. The pre-training includes the following specific steps:

[0135] Collect the training sample set , where is the group of training samples, [[ID= forty-two]]<、 、 and are respectively the ambient temperature, ambient humidity, ambient air pressure, and outlet water temperature in the the group of training samples , , is the total number of training samples;

[0136] Define the model as Using an insensitive loss function To measure the error between the predicted value and the actual value, the first Group training samples The error is When the error Less than or equal to the insensitive threshold Insensitive loss When the error is 0 Greater than the insensitive threshold Insensitive loss For error The absolute value minus the insensitive threshold ,in, Represents the input vector, the first... Group training samples input vector ;

[0137] Select kernel function As the radial basis kernel function, further determination is made based on the ambient temperature. Ambient humidity and ambient air pressure Transformation function mapping to higher-dimensional space Transformation function The inner product of the transformation function values ​​of two input vectors is equal to the kernel function values ​​of the two input vectors, as shown in the following formula:

[0138] ;

[0139] in, and The first Group training samples and the Group training samples The input vector;

[0140] The goal of support vector regression models is to find the optimal model representation. , and These are the optimal weight vector and the optimal bias term, respectively; that is, finding a set of weight vectors. and bias terms This makes the training sample set middle The optimization objective is to minimize the insensitive loss of each training sample, as follows:

[0141] ;

[0142] in, is a weight vector is a regularization term to avoid overfitting of the support vector regression model, is a penalty factor;

[0143] Since the insensitive loss function is a piecewise function with existing limits, the first group of training samples introduce a first Lagrange multiplier and a second Lagrange multiplier , a total of Lagrange multipliers, the optimization objective is rewritten as a Lagrange function , as follows:

[0144] ;

[0145] wherein, and are respectively a first multiplier vector and a second multiplier vector composed of the first Lagrange multipliers and the second Lagrange multipliers of the training samples;

[0146] Based on the dual theory, the partial derivative of the Lagrange function with respect to the weight vector and the bias term is taken and set to 0 to obtain a dual problem equivalent to the optimization objective, and a sequential minimal optimization algorithm is used to solve to obtain the optimal first multiplier vector and the optimal second multiplier vector ;

[0147] The optimal first multiplier vector and the optimal second multiplier vector are substituted back into the Lagrange function with respect to the weight vector and the bias term , and the partial derivative is set to 0 to solve to obtain the optimal weight vector and the optimal bias term , obtaining the optimal model representation , and the pre-training of the support vector regression model is completed, represents the transpose of the optimal weight vector .

[0148] Embodiment 2

[0149] As shown in Figure 3 , one specific embodiment of the present application discloses an intelligent water outlet control method for a water purification device, which is realized based on an intelligent water outlet control system for a water purification device, and includes the following specific steps:

[0150] Sensing load weight Change from 0 to non-zero to obtain the load-bearing weight. The laser rangefinder is activated to project a laser dot matrix to collect distance data points. The container capacity was estimated using a three-dimensional reconstruction volume estimation algorithm. According to the water output ratio Determine water output And calculate the expected weight Fuzzy inference is used to adaptively determine the water output velocity. ;

[0151] Set the waiting time, and the integrated voice device will continuously perform vibration sensing to capture sound waves that may appear in the environment and convert them into voice signals. ;

[0152] If a voice signal is obtained during the waiting time The preprocessed speech signal is analyzed using an autoencoder speech recognition model. Generate speech content Based on the trigger-effective strategy, an automaton algorithm combined with regular expression matching is used to determine the speech content. The validity of the speech content is if and only if the speech content is valid. The text contains both keywords and effluent temperature. At that time, determine the content of the speech. Effective, extraction water temperature Then turn off the integrated voice device; otherwise, continue to wait for a new voice signal. Until the waiting time ends;

[0153] If no voice signal is received within the waiting time Or not through voice signal Obtain the outlet water temperature The temperature sensor, humidity sensor, and barometric pressure sensor are activated to collect ambient temperature data. Ambient humidity and ambient air pressure And a support vector regression model combined with ambient temperature was used. Ambient humidity and ambient air pressure Predict and obtain suitable outlet water temperature ;

[0154] Based on water output speed Water output and outlet water temperature During water discharge, a weight gradient monitoring strategy is implemented to monitor the load-bearing weight in real time. And periodically determine the first-order gradient and second order gradient to induce whether the container is abnormal, immediately stop water when the container is abnormal, and play abnormal prompt voice through integrated voice device, and continue to water until the bearing weight is equal to the expected weight Stop water and play completion prompt voice through integrated voice device.

[0155] Example 3

[0156] As Figure 4 shown, one specific embodiment of the present application discloses a water purification device for carrying the intelligent water control system for water purification device, including water purification machine 1, gravity sensor 2, laser range finder 3, temperature sensor 4, humidity sensor 5, air pressure sensor 6, integrated voice device 7 and single-chip microcomputer 8.

[0157] Water purification machine 1 has heating function, removes impurities and bacteria in water through filtration technology, and provides drinking water under the control of single-chip microcomputer 8 with water speed , water volume and water temperature ;

[0158] Gravity sensor 2 is installed below the water receiving platform of water purification machine 1, continuously senses the bearing weight of water receiving platform , and shares the bearing weight wirelessly with single-chip microcomputer 8 when the bearing weight is not 0;

[0159] Laser range finder 3 is installed above the water receiving platform of water purification machine 1, controlled by single-chip microcomputer 8 to vertically project laser dot matrix to water receiving platform to feed distance data point set to single-chip microcomputer 8;

[0160] Temperature sensor 4 is installed on one side of water purification machine 1, controlled by single-chip microcomputer 8, obtains environmental temperature by measuring the forward voltage drop of semiconductor PN junction and feeds back to single-chip microcomputer 8;

[0161] Humidity sensor 5 is installed on one side of water purification machine 1, controlled by single-chip microcomputer 8, obtains environmental humidity by measuring the capacitance of humidity sensitive medium between parallel plate metal electrodes and feeds back to single-chip microcomputer 8;

[0162] Air pressure sensor 6 is installed on one side of water purification machine 1, controlled by single-chip microcomputer 8, obtains environmental air pressure by measuring the voltage signal output by Wheatstone bridge and feeds back to single-chip microcomputer 8;

[0163] The integrated voice device 7 is installed on one side of the water purifier 1, and includes a MEMS microphone 71 and a speaker 72; the MEMS microphone 71 is controlled to start by the single-chip microcomputer 8, and continuously performs vibration sensing to capture sound waves in the environment, and generates a voice signal The speaker 72 is controlled to play an abnormal prompt voice or a completion prompt voice by the single-chip microcomputer 8;

[0164] The single-chip microcomputer 8 is arranged in the water purifier 1, and is used for controlling the laser range finder 3, the temperature sensor 4, the humidity sensor 5, the air pressure sensor 6 and the integrated voice device 7 to perform collection or playing, and determining an outflow speed, an outflow amount and an outflow temperature based on feedback data and controlling the water purifier 1 to outflow water, judging a container abnormality by monitoring a bearing weight shared by the gravity sensor 2 and stopping the water purifier 1 to outflow water when the container abnormality or the outflow is completed.

[0165] The application discloses an intelligent outflow control system and method for a water purifying device and the water purifying device.

[0166] The above only describes preferred embodiments of the application, and the protection scope of the application is not limited to the above-described embodiments, and any technical scheme falling within the idea of the application belongs to the protection scope of the application. It should be noted that, for ordinary skilled persons in the art, some improvements and decorations without departing from the principle of the application are also regarded as the protection scope of the application.​​

Claims

1. An intelligent water outlet control system for a water purification apparatus, characterized in that, The central module receives the load weight, broadcasts the ranging signal and the activation signal, sets the waiting time, receives the distance data set, estimates the container capacity by using the three-dimensional reconstruction volume estimation algorithm, converts the container capacity into the water output in a proportional manner, determines the expected weight, determines the water output speed by fuzzy reasoning based on the container capacity, selects the self-encoding speech recognition model based on the reception of the voice signal within the waiting time, analyzes the effectiveness of the voice signal by combining the trigger effective strategy, extracts the water temperature, broadcasts the silence signal or directly broadcasts the sensing signal and the silence signal, receives the environmental data, predicts the water temperature by using the support vector regression model, performs the water output based on the water output speed, the water output and the water temperature, continuously monitors the load weight, and periodically calculates the first-order gradient and the second-order gradient of the load weight by using the weight gradient supervision strategy to sense the container state, stops the water output when the container is abnormal and broadcasts the abnormal signal, and continues the water output until the expected weight is reached and the completion signal is broadcasted when there is no abnormality. The central module includes a speech recognition unit.

2. The intelligent water outlet control system for water purification apparatus as claimed in claim 1, wherein, The speech recognition unit receives and pre-processes the voice signal to generate voice features, inputs the self-encoding speech recognition model, combines the local features extracted by the combined convolutional network with the global features extracted by the Transformer encoder to generate speech codes, uses the CTC decoder to assist the Transformer decoder to learn the alignment relationship between the speech codes and Chinese characters, generates the speech content, and extracts the water temperature based on the trigger effective strategy to identify the effective speech content. The generation of the speech content includes the following specific steps:

3. The intelligent water outlet control system for water purification apparatus as claimed in claim 2, wherein, The voice signal is divided into multiple segment signals by a first-order high-pass filter, the product of each segment signal and the Hamming window function is converted to the frequency domain by Fourier transform to generate the frequency spectrum information of the voice signal, a Mel filter bank is constructed based on the conversion relationship between the Mel frequency and the linear frequency, and the voice features are obtained by convolution with the frequency spectrum information and taking the logarithm; The trigger effective strategy is used to determine the effectiveness of the speech content, and the automatic machine algorithm is used to match the keywords in the speech content based on the keyword regular expression, if the keywords exist, it is determined that the speech content is effective, and the automatic machine algorithm is used to extract the water temperature in the speech content based on the temperature regular expression, if the keywords do not exist, it is determined that the speech content is invalid. A convolutional encoder of a self-encoding speech recognition model processes speech features to generate speech codes, the convolutional encoder combines an existing structure of a Transformer encoder with a combined convolutional network through a Makaron structure, the combined convolutional network includes a pointwise convolutional layer, a gated linear layer, a channel-wise convolutional layer and an activation function, wherein the pointwise convolutional layer is used to extract channel information, the gated linear layer is used to filter features, and the channel-wise convolutional layer is used to extract local feature information in the time dimension, and the activation function is used to increase the nonlinear expression capability; The joint encoder of the self-encoding speech recognition model introduces a CTC decoder in the Transformer decoder, which includes a linear layer and an activation function for establishing a probability distribution of the alignment path of the speech encoding to the Chinese character sequence, the Transformer decoder converts the speech encoding into Chinese character content and adjusts the Chinese character order based on the probability distribution of the alignment path to generate the speech content.

4. The intelligent water outlet control system for water purification apparatus as claimed in claim 2, wherein, The three-dimensional reconstruction volume estimation algorithm for estimating the container capacity includes the following specific steps:

5. The intelligent water outlet control system for water purification apparatus as claimed in claim 1, wherein, A space coordinate system is constructed and the distance data set is placed into it to generate a point cloud data set; The partial derivatives of a single target point in three coordinate axis directions are calculated to construct a gradient vector, if the gradient vector modulus value of the single target point is greater than or equal to the gradient threshold, it is determined as a container edge point, all container edge points are obtained, and the local surface of the non-edge points is removed to generate a preliminary container three-dimensional model; Traverse the point cloud dataset, through The nearest neighbor algorithm determines the neighborhood of a single target point and the nearest neighbor points within the neighborhood. The least squares method is used to fit the local surface with the smallest distance to all nearest neighbor points to obtain the local surface of all single target points and construct a preliminary three-dimensional model. The covariance matrix is calculated based on all the neighboring points of the single target point and the eigenvalue decomposition is performed to obtain three eigenvalues, the ratio of the minimum eigenvalue and the sum of the three eigenvalues is taken as the curvature of the single target point, and the curvature is compared with the curvature threshold, if the curvature is greater than or equal to the curvature threshold, it is determined as a corner point, and all corner points are obtained. ​ The Laplace smoothing algorithm is used to calculate the neighborhood average coordinates of all the neighboring points of each corner point, and the corner point is offset to the neighborhood average coordinates by a smoothing factor to smooth the surface of the preliminary container three-dimensional model, so as to generate the container three-dimensional model; The micro-element integration method is used to divide the container three-dimensional model into multiple cuboids, and the volume of all the cuboids is superimposed by double integration to estimate the container capacity, or a proportional calculation method is used to generate multiple particles in the minimum oriented bounding box of the container three-dimensional model and calculate the inner particle ratio, and the product of the volume of the minimum oriented bounding box and the inner particle ratio is taken as the container capacity, wherein the inner particle ratio is the ratio of the total number of inner particles located in the container three-dimensional model to the total number of particles.

6. The intelligent water outlet control system for water purification apparatus as claimed in claim 1, wherein, The support vector regression model needs to be pre-trained, and the pre-training includes the following specific steps: Collect multiple groups of training samples to construct a training sample set, and each group of training samples includes an input vector and an outlet water temperature, wherein the input vector includes an ambient temperature, an ambient humidity and an ambient pressure; Define a model representation, and use an insensitive loss function to measure the error, if the error of each group of training samples is less than or equal to an insensitive threshold, the insensitive loss is 0, if the error is greater than the insensitive threshold, the insensitive loss is the absolute value of the error minus the insensitive threshold, and the error is the difference between the predicted value obtained by substituting the input vector into the model representation and the outlet water temperature; Select a radial basis kernel function as a kernel function and determine a transformation function for mapping the input vector to a high-dimensional space, and the transformation function satisfies that the inner product of the transformation function values of two input vectors is equal to the kernel function value of the two input vectors; The target of the support vector regression model is to find an optimal model representation, which is a linear expression with respect to an optimal weight vector and an optimal bias term, that is, to find a group of weight vectors and bias terms, so that the insensitive loss of all training samples is minimized, and a regularization term of the weight vector is introduced to construct an optimization target; Introduce a Lagrange multiplier to rewrite the optimization target into a Lagrange function, wherein the variables of the Lagrange function include the weight vector, the bias term, a first multiplier vector composed of first Lagrange multipliers of all training samples and a second multiplier vector composed of second Lagrange multipliers of all training samples; Take the partial derivative of the Lagrange function with respect to the weight vector and the bias term and set it equal to 0 to obtain a dual problem and solve it by using a sequential minimal optimization algorithm to obtain an optimal first multiplier vector and an optimal second multiplier vector, and then substitute the optimal first multiplier vector and the optimal second multiplier vector into the partial derivative of the Lagrange function with respect to the weight vector and the bias term, and set the partial derivative equal to 0 to solve and obtain an optimal weight vector and an optimal bias term, and the pre-training is completed.

7. The intelligent water outlet control system for water purification apparatus as claimed in claim 4, wherein, The automatic machine algorithm uses a deterministic finite automaton, defines a finite state set based on a regular expression, includes an initial state and an accepting state, respectively corresponding to an empty string and the regular expression, starts from the initial state, reads the voice content character by character, inputs the current state and the current read character into a state transition function to generate the next state, and after completely reading the voice content, if the deterministic finite automaton is in the accepting state, the voice content matches the regular expression successfully, and in other cases, the matching fails.

8. The intelligent water outlet control system for water purification apparatus as claimed in claim 1, wherein, The system further comprises a weight sensing module, a comprehensive acquisition module and a voice module; The weight sensing module shares the bearing weight with the central module when the bearing weight is not 0; The comprehensive acquisition module receives ranging signals or sensing signals, acquires distance data point sets or acquires environmental data, and the environmental data includes environmental temperature, environmental humidity and environmental air pressure; The voice module executes vibration sensing to generate voice signals when receiving an activation signal, gives voice prompts when receiving an abnormal signal or a completion signal, and stops vibration sensing when receiving a silence signal.

9. A method for intelligent water outlet control for a water purification apparatus, characterized by, The specific steps include: Obtaining a bearing weight that is not 0, acquiring distance data point sets and estimating the volume of the container by a three-dimensional reconstruction volume estimation algorithm, determining the water output according to the water output ratio and calculating the expected weight, and determining the water output speed based on fuzzy reasoning; Setting a waiting time, executing vibration sensing to capture sound waves in the environment and converting them into voice signals; If voice signals are obtained within the waiting time, a self-encoding voice recognition model is used to analyze the preprocessed voice signals to generate voice content, based on a trigger validity strategy, an automatic machine algorithm is used to judge the validity of the voice content by combining regular expressions, when the voice content is valid, the water output temperature is extracted and the vibration sensing is stopped, and when the voice content is invalid, new voice signals are continuously waited until the waiting time ends; If no voice signals are obtained within the waiting time or all the obtained voice signals are invalid, the environmental temperature, humidity and air pressure are collected, and the water output temperature is predicted by a support vector regression model; According to the water output speed, the water output and the water output temperature, the water output and the weight gradient supervision strategy are executed, the bearing weight is monitored in real time, and the volatility of the first-order gradient and the second-order gradient is periodically judged to sense whether the container is abnormal, the water output is stopped and an abnormal prompt voice is played when the container is abnormal, and the water output continues until the bearing weight equals the expected weight and the water output is stopped and a completion prompt voice is played when the container is normal.

10. A water purification apparatus for carrying an intelligent water outlet control system for the water purification apparatus and performing an intelligent water outlet control method for the water purification apparatus, characterized by, The system comprises a gravity sensor, a laser range finder, a temperature sensor, a humidity sensor, an air pressure sensor, an integrated voice device and a single-chip microcomputer; The gravity sensor continuously senses the bearing weight of the water receiving platform, and shares the bearing weight with the single-chip microcomputer when the bearing weight is not 0; The laser range finder is controlled by the single-chip microcomputer to vertically project a laser dot array to the water receiving platform to feed back distance data point sets to the single-chip microcomputer; The temperature sensor is controlled by the single-chip microcomputer to obtain the environmental temperature and feed back to the single-chip microcomputer; The humidity sensor is controlled by the single-chip microcomputer to obtain the environmental humidity and feed back to the single-chip microcomputer; The air pressure sensor is controlled by the single-chip microcomputer to obtain the environmental air pressure and feed back to the single-chip microcomputer; The integrated voice device is controlled by the single-chip microcomputer to execute vibration sensing to capture sound waves in the environment, generate voice signals and feed back to the single-chip microcomputer or play abnormal prompt voice and completion prompt voice; The single-chip microcomputer controls the laser range finder, the temperature sensor, the humidity sensor, the air pressure sensor and the integrated voice device to execute acquisition or play, determines the water output speed, the water output and the water output temperature based on the feedback data, and controls the water purifier to output water, judges the container abnormality by monitoring the bearing weight shared by the gravity sensor, and stops the water output when the container is abnormal or the water output is completed.

Images